Relationship Between Health Care Costs and Very Low Literacy Skills in a Medically Needy and Indigent Medicaid Population

Subjects

Subjects were selected using computer-generated random numbers from among enrollees in what was, at the time of the study in 1992, the largest health plan in southern Arizona providing care to enrollees in Arizona’s managed-Medicaid plan, the Arizona Health Care Cost Containment System (AHCCCS). Subjects had to be ≥18 years old, speak English or Spanish, and be enrolled continuously in AHCCCS through MNMI eligibility for at least 1 year before study.

Subject Interviews and Literacy Assessments

All subjects in this study gave informed consent. The written informed-consent process included use of a specially designed low-literacy consent form. Subjects also received an oral explanation of the study and its methods, which were approved by the University of Arizona Human Subjects Protection Program.

Each subject underwent a literacy skills test using the Instrument for the Diagnosis of Reading (IDR),8–11 which is an English-Spanish instrument; subjects were tested in their preferred language. IDR yields scores from 0 to 8, corresponding to grade-equivalent reading levels; a score of 8 indicates ≥8th-grade level.

We categorized subjects with IDR scores ≤3rd-grade level as having low reading skills and those with scores ≥4rd-grade level as having higher-level skills. There were 2 reasons for selecting this cutoff level. First, prior studies indicated that health knowledge and health status indicators are particularly poor for subjects with reading skills at or below the 2nd- or 3rd-grade level.2–7,9 Second, the most widely used health literacy assessment instruments categorizes subjects as being in the lowest level of literacy if reading skills are ≤3rd-grade level.12,13

Measuring Health Care Charges

We used health plan billing records to tabulate charges for subjects’ medical care, which included hospital, emergency department, short-term nursing home, and physician care, plus outpatient and inpatient charges for laboratory, radiographs, pharmacy, and durable medical equipment. These were summed to yield “total charges.”

Demographic Variables

During the interviews, we collected information on each subject’s age, self-reported ethnic group, health status, and education level. Health status was measured by asking subjects for a self-assessment of their health (excellent, good, fair, poor).

Data Analysis

Using SAS software,14 we performed t tests to measure differences in health care costs between subjects with low- and high-level reading groups, as defined above. We then performed a multivariable analysis was to determine whether literacy was a predictor of costs after considering age, ethnic group, health status, and education level.

Subjects

There were 74 MNMI subjects, representing 18% of the original 402 subjects, the remainder of whom were enrolled in the Medicaid program through eligibility criteria other than MNMI. The subjects ranged in age from 21 to 77 (mean age of 49.9 ± 13.7) and had completed a mean ± S.D. of 9.1 ± 4.0 years of education, with a range of 0 to 13. Table 1 displays other demographic characteristics of subjects.

Reading Levels and Charges for Health Care

The subjects’ mean reading level was at grade 5.4 (±2.7). Eighteen (24%) had IDR scores at low-literacy levels (≤3rd-grade) and 56 (76%) at higher levels. The 18 low-level readers generated significantly higher annual mean total health care charges than the 56 higher level readers ($10,688 vs $2,891; P = .025). This difference was caused principally by differences in inpatient care charges ($7,038 vs $824, P = .020) (Table 2).

The multivariable analysis revealed that reading level was still a significant predictor of costs, even after considering age, ethnic group, and health status (P = .037). In a second analysis that excluded reading level but retained education level, education level was not a significant predictor of health care charges (P = .09).